Yuki Takezawa
@yukitakeza
3rd year Ph.D. student @ Kyoto Univ.
ID: 1690636466316275713
https://yukitakezawa.github.io 13-08-2023 08:08:59
33 Tweet
111 Followers
196 Following
OISTの機械学習とデータ科学ユニットからICLR2025に5本論文が採択されました🎉🎉🎉 Yuki Takezawa Kira Düsterwald Satoki I Mohammad Sabokrou, Cindy おめでとうございます！！！ #ICLR2025 oist.mlds.jp/2025/01/23/fiv…
🧗♂️Why GD converges beyond [step size]<2/[smoothness]? We investigate loss functions and identify their *separation margin* is an important factor. Surprisingly Renyi 2-entropy yields super fast rate T=Ω(ε^{-1/3})! (collab w/ Shinsaku Sakaue Yuki Takezawa) arxiv.org/abs/2502.04889
![🧗♂️Why GD converges beyond [step size]<2/[smoothness]? We investigate loss functions and identify their *separation margin* is an important factor. Surprisingly Renyi 2-entropy yields super fast rate T=Ω(ε^{-1/3})!
(collab w/ <a href="/shinsaku_sakaue/">Shinsaku Sakaue</a> <a href="/YukiTakeza/">Yuki Takezawa</a>)
arxiv.org/abs/2502.04889](https://pbs.twimg.com/media/GjZO9b6awAAT88V.png "HB (@levelfour_) on Twitter photo 🧗♂️Why GD converges beyond [step size]<2/[smoothness]? We investigate loss functions and identify their *separation margin* is an important factor. Surprisingly Renyi 2-entropy yields super fast rate T=Ω(ε^{-1/3})!
(collab w/ <a href=\"/shinsaku_sakaue/\">Shinsaku Sakaue</a> <a href=\"/YukiTakeza/\">Yuki Takezawa</a>)
arxiv.org/abs/2502.04889")
